Automatic Processing of EEG-EOG-EMG Artifacts in Sleep Stage Classification

In this paper, we present a series of algorithms for dealing with artifacts in electroencephalograms (EEG), electrooculograms (EOG) and electromyograms (EMG). The aim is to apply artifact correction whenever possible in order to lose a minimum of data, and to identify the remaining artifacts so as not take them into account during the sleep stage classification. Nine procedures were implemented to minimize cardiac interference and slow ondulations, and to detect muscle artifacts, failing electrode, 50/60Hz main interference, saturations, highlights abrupt transitions, EOG interferences and artifacts in EOG. Detection methods were developed in the time domain as well as in the frequency domain, using adjustable parameters. A database of 20 excerpts of polysomnographic sleep recordings scored in artifacts by an expert was available for developing (excerpts 1 to 10) and testing (excerpts 11 to 20) the automatic artifact detection algorithms. We obtained a global agreement rate of 96.06%, with sensitivity and specificity of 83.67% and 96.47% respectively. Keywords— Artifacts processing, EEG, EOG, EMG, ECG.